Causalnex Versions Save

Release 0.12.1

Release 0.12.0

Release 0.11.2

Release 0.11.1

Change log:

• Add python 3.9, 3.10 support
• Unlock Scipy restrictions
• Fix bug: infinite loop on lv inference engine
• Fix DAGLayer moving out of gpu during optimization step of Pytorch learning
• Fix CPD comparison of floating point - rounding issue
• Fix set_cpd for parentless nodes that are not MultiIndex
• Add Docker files for development on a dockerized environment

Release 0.11.0

Changelog:

• Add expectation-maximisation (EM) algorithm to learn with latent variables
• Add a new tutorial on adding latent variable as well as identifying its candidate location
• Allow users to provide self-defined CPD, as per #18 and #99
• Generalise the utility function to get Markov blanket and incorporate it within StructureModel (cf. #136)
• Add a link to PyGraphviz installation guide under the installation prerequisites
• Add GPU support to Pytorch implementation, as requested in #56 and #114 (some issues remain)
• Add an example for structure model exporting into first causalnex tutorial, as per #124 and #129
• Fix infinite loop when querying InferenceEngine after a do-intervention that splits the graph into two or more subgraphs, as per #45 and #100
• Fix decision tree and mdlp discretisations bug when input data is shuffled
• Fix broken URLs in FAQ documentation, as per #113 and #125
• Fix integer index type checking for timeseries data, as per #74 and #86
• Fix bug where inputs to the DAGRegressor/Classifier yielded different predictions between float and int dtypes, as per #140

Functionality:

• Add BayesianNetworkClassifier an sklearn compatible class for fitting and predicting probabilities in a BN.
• Add supervised discretisation strategies using Decision Tree and MDLP algorithms.
• Support receiving a list of inputs for InferenceEngine with a multiprocessing option
• Add utility function to extract Markov blanket from a Bayesian Network

Minor fixes and housekeeping:

• Fix estimator issues with sklearn ("unofficial python 3.9 support", doesn't work with discretiser option)
• Fixes cyclical import of causalnex.plots, as per #106.
• Added manifest files to ensure requirements and licenses are packaged
• Minor bumps in dependency versions, remove prettytable as dependency

No functional changes.

Docs:

• Remove Boston housing dataset from the "sklearn tutorial", see #91 for more information.

Development experience:

• Update pylint version to 2.7
• Improve speed and non-stochasticity of tests

• Fixed bug where the sklearn tutorial documentation wasn't rendering.
• Weaken pandas requirements to >=1.0, <2.0 (was ~=1.1).